← Back to Sessions

Earthquake Rate Forecasting in New Zealand Using Multiplicative Hybrid Models Including Strain-Rate Covariates: Performance Analysis in Forward and Reverse Time

Session: Beyond Poisson: Seismic Hazards and Risk Assessment for the Real Earth [Poster]

Type: Poster

Date: 4/19/2021

Presentation Time: 11:30 AM Pacific

Description: 

Long term earthquake rate forecasting is an essential part of the National Seismic Hazard Model of New Zealand (NZNSHM). Multiplicative hybrid models including strain rates by Beavan (2012) and other contributing covariates based on fault and earthquake data were previously fitted to NZ earthquakes with M ≥ 5.0 over 1987 to 2006. Such models serve to improve the long term seismicity rate forecast for the NZNSHM. The strain rate covariates were based on geodetic data from 1991-2011. Maximum shear strain rate was the most informative of the covariates and all hybrid models including strain rates were more informative than models excluding them. The same was true in forward testing from 2012-2015 – a period independent of data contributing to strain rate estimates. We incorporate an updated shear strain rate model (Haines and Wallace 2020) based on geodetic data up to 2019 and obtain multiplicative rate models using the same earthquakes and fitting period. In forward testing hybrids with Haines and Wallace (2020) shear strain rate outperform all others. We investigate information gains for the models when performing reverse testing on an independent period 1951-1986. All hybrid models including Beavan (2012) and Haines and Wallace (2020) perform poorly over this period and are less informative than hybrids excluding strain rates. We evaluate competency of covariates as predictors using Molchan error diagrams and area skill score (ASS). Although the shear strain rates are positively correlated with earthquake occurrence over the reverse period, other covariates like proximity to mapped faults have higher ASS and are better correlated. Smoothed scatter plots of covariates for targeted earthquakes against time 1951-2020 show the association between shear strain and earthquake occurrence is strongest during the time on which the strain rate model was based. These results are relevant to question of how strongly strain rate estimates based on a few decades of geodetic data should be weighted when constructing earthquake source models for assessment of long term seismic hazard.

Presenting Author: Sepideh J Rastin

Student Presenter: No

Authors